[FEATURE] Prompt Queue & Steering for Auto-Multiturn Agent Execution

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Feature: Prompt Queue & Steering for Auto-Multiturn Agent Execution

Summary

Add a Prompt Queue & Steering system that allows the OpenCode agent to execute multiple prompts sequentially in an ordered queue, enabling true auto-multiturn workflows without requiring manual intervention for each step.

Problem

Currently, the agent requires manual sending for each prompt/turn. For long-running coding tasks (e.g., scaffolding a project, running tests, fixing errors iteratively), users have to wait and manually trigger each step. This breaks flow and makes it impossible to "set it and forget it."

Proposed Solution

1. Prompt Queue

• Allow users to enqueue multiple prompts/tasks in a specific order
• Agent processes the queue step-by-step, one prompt at a time
• Queue persists across turns — user can close the session and come back later
• Visual queue UI showing pending, active, and completed items

2. Steering

• Each queued prompt can have steering parameters:
• • max_iterations: how many retry/self-correct loops before moving on
  • stop_condition: what success looks like (tests pass, build succeeds, etc.)
  • on_error: skip / retry / pause for human input
• Agent auto-adjusts based on steering instead of asking for permission at every step

3. Direct Send Override

• Keep the existing "Send" button for instant execution (no queue)
• Add a "Queue & Send Later" or "Append to Queue" option
• Toggle between direct mode and queued mode per session

Example Workflow

User queues:
1. "Scaffold a React + TypeScript + Vite app"
2. 
2. "Install dependencies and run npm run dev"
3. 
3. "If it fails, debug and fix"
4. 
4. "Add Jest testing setup"
5. 
5. "Write 5 unit tests for the main component"
6. 

Agent executes 1→5 sequentially, handling failures according to steering rules.
User can walk away. Results summarized at the end.

Benefits

• Hands-off coding: Queue a full task list, let the agent work through it
• Reproducible workflows: Same queue = same execution order
• Better for CI-like tasks: Agent can run tests, lint, fix, re-run
• Flexibility: Direct send still available when you want instant control

Related

• Similar to Task Queues in LangChain / AutoGen
• Inspired by "Set it and forget it" agentic workflows in Devika, OpenDevin
• Would pair well with a background agent mode / daemon process

---

Would love feedback on the API design and whether this fits the OpenCode architecture!